Machine Learning Based WIFI Indoor Positioning Technology Research

Location-based services(LBS)are widely used in many fields,showing huge commercial potential,and the positioning technology,which serves as its core support,has always been a research hotspot in industry and academia.In the field of indoor positioning,the universal positioning technology with high accuracy,low complex and low-cost is the goal of people's research and the direction of future development.WIFI fingerprint positioning does not require additional special equipment,and has the advantages of low cost and universality.What's more,considering that WIFI fingerprint positioning can be modeled as a nonlinear supervised learning problem,the WIFI indoor positioning technology based on machine learning has important research significance.Firstly,by combining the measured RSS data of WIFI signal,it is found that there exists time-varying nonlinear characteristic between the indoor location and the RSS fingerprint of the WIFI signal,and the mechanism behind the location fingerprint positioning technology is investigated.Then,in order to improve positioning accuracy,this paper focus on studying which machine learning algorithm is more suitable for fitting the time-varying nonlinear relationship between the location and the RSS fingerprint of WIFI signal.After respective regression and classification positioning algorithm models are established for the commonly used SVM algorithm and bagging algorithm Random Forest(RF)in ensemble learning,a new boosting algorithm Gradient Boosting Decision Tree(GBDT)in ensemble learning is introduced and its positioning algorithm model is established.These positioning algorithms are simulated on the location fingerprint database established by the ray tracing method in the simulated environment.After the comprehensive comparison of relevant indicators is completed,the positioning algorithm with the best performance is fixed.The experimental results show that the positioning accuracy obtained by using the regression algorithm based on the same machine learning algorithm is higher.In the simulated environment,the introduced GBDT positioning algorithm has the better positioning performance.For the positioning of moving targets,in order to make full use of the prior information and further improve the positioning accuracy,a location tracking algorithm that combines machine learning and filtering methods is proposed.Specifically,fusion location tracking algorithms model based on GBDT-KF and GBDT-PF are established,the location tracking simulation of two fusion algorithms on three types of paths in the simulation environment are finished.The experimental results show that the positioning accuracy of the proposed fusion positioning algorithms GBDT-KF and GBDT-PF are at least 30%higher than that of the positioning algorithm based on GBDT,and the fusion positioning algorithms GBDT-KF and GBDT-PF are more suitable for linear and nonlinear positioning tracking paths in the simulated environment,respectively.Finally,by combining the measured data in the actual scene,feasibility of the introduced and proposed algorithms is verified.The experimental results show that,in the actual environment,the positioning performance of the introduced GBDT positioning algorithm based on ensemble learning has high accuracy which reaches 1.02m in average,the proposed fusion positioning algorithms have higher positioning accuracy than GBDT positioning algorithm,and its location tracking results are also closer to the actual trajectory,which further verify the feasibility of the introduced and proposed algorithms.